Job Listing Evaluation Skill

Evaluate listings after retrieval.

Follow this workflow

1. Read the candidate profile first.
2. Read the collected listing.
3. Score the listing on a single 0-100 scale.
4. Write exactly one JSON artifact per listing to the caller-provided output path.
5. Keep reasoning concise and specific.
6. Make hard mismatches explicit; do not hide them behind a middling score.

Decision rules

• Use keep when the listing is plausibly worth human review.
• Use drop when the listing has a clear seniority, experience, location, work-mode, authorization, or role-family mismatch.
• A high score with a drop decision is not allowed.
• If critical evidence is missing, say so in the reasoning instead of inventing assumptions.

Output contract

Use the schema in references/evaluation-schema.md.

Write one file per listing to:

• runtime-data/evaluations/<runId>/<listingId>.json

If the evaluator cannot complete after reading inputs, it may write:

• runtime-data/evaluations/<runId>/<listingId>.error.json

Do not rely on stdout to carry the evaluation payload. Stdout should only confirm success or report failure.

If the user wants an example payload or field meanings, read that reference before responding.